Back to AIBriefs
AnalysisAI Models

Spatial vision-language models consistently wrong on distance queries

Arxiv CS.CV (computer vision)

Researchers find VLMs produce consistent but incorrect metric distance predictions across viewpoints, indicating evidence insensitivity. The failure undermines reliability for spatial tasks in robotics and embodied AI.

·
8 days ago
Spatial vision-language models consistently wrong on distance queries — AIBriefs